Evaluation of GPU-Accelerated Edge Platforms for Stochastic Simulations: Performance and Energy Efficiency Analysis

With the increasing emphasis on energy-efficient computing, edge devices accelerated by graphics processing units (GPUs) are gaining attention for their potential in scientific workloads. These platforms support compute-intensive simulations under strict energy and resource constraints, yet their computational efficiency across architectures remains an open question. This study evaluates the performance of GPU-based edge platforms for executing the stochastic simulation algorithm (SSA), a widely used and inherently compute-intensive method for modeling biochemical and physical systems. Execution time, floating point throughput, and the trade-offs between cost and power consumption are analyzed, with a focus on how variations in core count, clock speed, and architectural features impact SSA scalability. Experimental results show that the Jetson Orin NX consistently outperforms Xavier NX and Orin Nano in both speed and efficiency, reaching up to 4.86 million iterations per second while operating under a 20 W power envelope. At the largest workload scale, it achieves 2102.7 ms/W in energy efficiency and 105.3 ms/USD in cost-performance—substantially better than the other Jetson devices. These findings highlight the architectural considerations necessary for selecting edge GPUs for scientific computing and offer practical guidance for deploying compute-intensive workloads beyond artificial intelligence (AI) applications.